DiSCo v2: the CUDA re-walk, every tier, scanner-class certificate

LICENSING.md

@@ -1,30 +0,0 @@
-# Licensing
-
-The replay packs of this dataset are dual-licensed.
-
-## Research and non-profit use — CC BY-NC-SA 4.0
-
-The packs, their manifest and their certificates are available under the
-[Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License](https://creativecommons.org/licenses/by-nc-sa/4.0/).
-You may download, replay, redistribute and build upon them for research, teaching and any other non-commercial
-purpose, as long as you credit this dataset and its source geometry (the `citation` field of every pack) and
-release what you build upon them under the same terms.
-
-## Commercial use
-
-A commercial licence is available under separately negotiated terms for use that makes money: inside a product,
-a paid service, contract research, or the training of a model that is sold. It lifts the non-commercial and
-share-alike obligations.
-
-📧 **rutger.fick@dmrai-lab.org**
-
-## What the licence covers, and what it does not
-
-The licence covers the simulated records in the packs: the walkers' paths, occupancies, wall contacts and field
-samples, their coefficients, and the certificates. The **source geometry** each pack was simulated on keeps its
-own licence, recorded in the pack's `license` field and in the manifest, and this dataset never relicenses it.
-The replay-pack **format** is a separate, openly licensed specification (CC-BY-4.0 text, Apache-2.0 reference code):
-anyone may produce and read packs under their own terms.
-
-Copies of a pack obtained under an earlier licence of this dataset keep that licence; these terms apply from the
-revision that carries this file.

README.md

@@ -1,7 +1,5 @@
 ---
-license:
-  - cc-by-nc-sa-4.0
-  - other
+license: cc-by-nc-sa-4.0
 pretty_name: DiSCo replay pack
 tags:
   - diffusion-mri
@@ -92,7 +90,7 @@ clinical scanner reads only a prefix of the bands.
 
 ## How it was made
 
-- **Walk:** dmipy-sim. Blocks of 99.5 µs with adaptive sub-steps, own-tube
+- **Walk:** the CUDA backend of dmipy-sim (commit `3b1207b`). Blocks of 99.5 µs with adaptive sub-steps, own-tube
   wall interaction, and the strand field read every 33 µs. The tiers were accumulated in the walk as band
   projections; no trajectory was stored.
 - **Planning:** walker counts per voxel and pool from a pilot at the class envelope, so that each voxel reaches the
@@ -100,6 +98,8 @@ clinical scanner reads only a prefix of the bands.
 - **Assembly:** 704 shards, merged in block order into one pack.
 - **Checks:** the merged pack was replayed against the time-segment layout at 16 voxels and the seven timing
   classes. The worst disagreement was 2.2 × 10⁻⁵, against a combined bound of 3.0 × 10⁻³.
+- **Earlier version:** the previous revision of this dataset holds the first DiSCo walk, a JAX walk on a save grid
+  with fewer certified tiers. It stays in the history.
 
 ## Licence